The rapid development of large language model techniques in recent years has made effective summarization of cross-lingual dialogue information possible, which is crucial in today’s global communication landscape. However, existing approaches often face problems with the lack of interpretability information and intermediate result analysis for the summarization generation process. In this paper, we propose two optimizations to address these issues. First, we use a self-reply analysis structure to extract the subtle attitude changes for each participant through the dialogue progress. Second, we combine this information to generate more interpretability cross-lingual dialogue summarization results. We propose a view-aware, chain-of-thought-based structure to clarify the generation process of cross-lingual dialogue summarization. The temporal properties of dialogue applications are considered throughout the computational process within our framework. Experimental results on cross-lingual summarization tasks in English, French, Spanish, Russian, Chinese, and Arabic, as well as cross-lingual hybrid tasks, demonstrate that our proposed method outperforms state-of-the-art baselines.